Kazuki Kozuka

Transfer learning method using multi-prediction deep Boltzmann machines for a small scale dataset

In this article, we propose a transfer learning method using the multi-prediction deep Boltzmann machine (MPDBM). In recent years, deep learning has been widely used in many applications such as image classification and object detection. However, it is hard to apply a deep learning method to medical images because the deep learning method needs a large number of training data to train the deep neural network. Medical image datasets such as X-ray CT image datasets do not have enough training data because of privacy. In this article, we propose a method that re-uses the network trained on non-medical images (source domain) to improve performance even if we have a small number of medical images (target domain). Our proposed method firstly trains the deep neural network for solving the source task using the MPDBM. Secondly, we evaluate the relation between the source domain and the target domain. To evaluate the relation, we input the target domain into the deep neural network trained on the source domain. Then, we compute the histograms based on the response of the output layer. After computing the histograms, we select the variables of the output layer corresponding to the target domain. Then, we tune the parameters in such a way that the selected variables respond as the outputs of the target domain. In this article, we use the MNIST dataset as the source domain and the lung dataset of the X-ray CT images as the target domain. Experimental results show that our proposed method can improve classification performance.

Determination of lung regions on chest ct images with diffuse lung diseases by use of anatomical structures and pulmonary textures.

To analyze diffuse lung diseases based on chest region computed tomography (CT) imaging by using a computer-aided diagnosis (CAD) system, it is necessary to first determine the lung regions subject to analysis. The lung regions can be selected relatively easily for healthy individuals, by applying a threshold. Selecting an area by using a threshold-based method can be difficult when dealing with lungs with diffuse lung diseases, owing to the abnormal opacities that characterize the diseases. Trials for determining the lung regions were conducted in this study, through texture analysis and machine learning, by narrowing down the lung regions to rough regions, and by referring to ribs and the diaphragm. This method can be used for determining lung regions for analysis of diffuse lung diseases.

A Proposed Speech Discrimination Assessment Methodology Based on Event-Related Potentials to Visual Stimuli

Objective and quantitative assessment methods are needed for the fitting of hearing aid parameters. This paper proposes a novel speech discrimination assessment method using electroencephalograms EEGs. The method utilizes event-related potentials ERPs to visual stimuli instead of the conventionally used auditory stimuli. A spoken letter is played through a speaker as an initial auditory stimulus. The same letter can then be visually displayed on a screen as a match condition, or a different letter is displayed mismatch condition. The participant determines whether the two stimuli represent the same letter or not. The P3 component or late positive potential LPP component are elicited when a participant detects either a match or mismatch between the auditory and visual stimuli, respectively. The hearing ability of each participant can be estimated objectively via analysis of these ERP components.

Learning Multiple Complex Features Based on Classification Results

Recently, methods for the unsupervised learning of features from large data sets have been attracting much attention. These methods have been especially successful in the area of computer vision. However, there is a problem that it is difficult to determine what kind of features will result in a high classification performance. Indeed, the difficulty of determining the learning parameters is a widely known problem in the field of feature learning. To address this problem, this paper presents a feature-learning method which uses classification results to progressively learn multiple features of varied complexity. The proposed method enables the learning of both simple robust features and complex features which represents difficult patterns. In addition, we assign regularization weights that are based on the complexity of the features. This modification emphasizes simple representation and prevents over fitting. Experimental results with medical image classification show that the proposed method is superior to the conventional method, especially when classification is difficult.

Assessment of Speech Discrimination Based on the Event-Related Potentials to the Visual Stimuli

Monitoring user states is one of the main applications in the e-Health field. This paper focuses on the assessment of speech discrimination for fitting the parameters of hearing aids using the electroencephalogram (EEG). The characteristic of our method is to utilize the event-related potentials (ERPs) to visual stimuli instead of the conventional auditory stimuli. A letter is played through a speaker as an initial auditory stimulus; then, the same letter is displayed on a screen as a following visual stimulus as a match condition (p = .50), and sometimes a different letter is displayed as a mismatch condition (p = .50). The participant then determines whether the two stimuli represent the same letter or not. The conventional evaluation focused on the evoked potentials just after the initial auditory stimuli. A larger component of late positive potential (LPP) is elicited when a participant detects the mismatch between the auditory and visual stimuli. These results suggest the possibility of assessing speech discrimination using EEG potentials.